We use cookies to ensure you get the best experience on our website
Halaman utama Jelajahi Bantuan News
Daftar Masuk
doi
/
Dataset_Levels
ter-fork dari fborn/Dataset_Levels
DOI
10.12751/g-node.hg4tdz
1
0
Fork 0
Berkas
Cabang: master
Ranting Tag
Dataset_Levels
/
processed_data
Frieda Born 5d7ff1b112 first commit of ressources for levels data repository 1 Minggu lalu
..
README.md 5d7ff1b112 first commit of ressources for levels data repository 1 Minggu lalu
processed_dataset.pkl 5d7ff1b112 first commit of ressources for levels data repository 1 Minggu lalu
pruned_dataset.pkl 5d7ff1b112 first commit of ressources for levels data repository 1 Minggu lalu
pruned_processed_dataset.pkl 5d7ff1b112 first commit of ressources for levels data repository 1 Minggu lalu

README.md

The processedLevels dataset

These files contain processed_data from the Levels dataset.

Overview

We have performed a couple of processing steps (e.g., removing duplicates, etc.) to give you the option to easily work with the data without having to process the sourcedata first. This resulted in 2 processed versions of the dataset:

  • pruned_processed_dataset.pkl: This file is processed and pruned, which means it only includes the experiment trials per participant and we already pruned to remove triplet responses of triplets that either have < 4 or > 5 responses.
  • processed_dataset.pkl: This dataset is processed (but not pruned). It includes only the necessary variables for analysis, but you would still need to filter for the experiment trials (see codebook) and please note, that here also duplicate triplet information is still included.

Loading data

If you use python, you can consider loading the data like this:

with open("../your path to data /pruned_processed_dataset.pkl", "rb") as f:
    pruned_processed_dataset = pickle.load(f)
datacite.yml
Title The Levels Dataset
Authors Muttenthaler,Lukas;Google DeepMind, Machine Learning Group, Technische Universität Berlin, BIFOLD, Berlin Institute for the Foundations of Learning and Data, Berlin, Germany;ORCID:0000-0002-0804-4687
Greff,Klaus;Google DeepMind;ORCID:0000-0001-6982-0937
Born,Frieda;Technische Universität Berlin, BIFOLD, Berlin Institute for the Foundations of Learning and Data, Berlin, Germany, Adaptive Memory and Decision Making (AMD), Max Planck Institute for Human Development,Berlin, Germany;ORCID:0009-0002-1214-4864
Spitzer,Bernhard;Adaptive Memory and Decision Making (AMD), Max Planck Institute for Human Development, Berlin, Germany;ORCID:0000-0001-9752-932X
Kornblith,Simon;Anthropic;ORCID:0000-0002-9088-2443
Mozer,Michael C.;Google DeepMind;ORCID:0000-0002-9654-0575
Müller,Klaus-Robert;Google DeepMind, Machine Learning Group, Technische Universität Berlin, BIFOLD, Berlin Institute for the Foundations of Learning and Data, Berlin, Germany,Department of Artificial Intelligence, Korea University, Seoul, Max Planck Institute for Informatics, Saarbrücken, Germany;ORCID:0000-0002-3861-7685
Unterthiner,Thomas;Google DeepMind;ORCID:0000-0001-5361-3087
Lampinen,Andrew K.;Google DeepMind;ORCID:0000-0002-6988-8437
Description To validate that AligNet can indeed help to increase the alignment between models and humans, we used crowd-sourcing to collect a novel evaluation dataset of human semantic judgments across multiple levels of abstraction that we call Levels.
License Open Data Commons Public Domain Dedication and License (PDDL) v1.0 (https://opendatacommons.org/licenses/pddl/1-0/)
References Muttenthaler, L., Greff, K., Born, F., Spitzer, B., Kornblith, S., Mozer, M.C., Müller, K.R., Unterthiner, T., Lampinen, A.K. : Aligning Machine and Human Visual Representations across Abstraction Levels [doi:10.48550/arXiv.2409.06509] (IsSupplementTo)
Born Frieda. (2024). Levels Collection Experiment Code (v1.0.0) [doi:10.5281/zenodo.13749102] (IsReferencedBy)
Funding
Keywords AI alignment
human cognition
representation learning
computer vision
Resource Type Dataset